Hazard Assessment Modeling: An Evolutionary Ensemble Approach

This paper presents a novel and effective genetic algorithm approach for generating computational models for hazard assessment. With millions of proposed chemicals being registered each year, it is impossible to come even remotely close to completing the battery of tests needed for the proper understanding of the toxic effects of these chemicals. Computer models can give quick, cheap, and environmentally friendly hazard assessments of chemicals. Our approach works by first extracting a hierarchy of theoretical descriptors of the structure of a compound, then filtering these numerous descriptors with a genetic algorithm approach to ensemble feature selection. We tested the utility of our approach by modeling the acute aquatic toxicity (LC50) of a congeneric set of 69 benzene derivatives. Our results demonstrate a very important point: that our method is able to accurately predict toxicity directly from structure.